At present, compression schemes for multi-channel audio signals do not explicitly take into account how the input audio material has been generated or mixed. Thus, known audio compression technologies are not aware of the origin/mixing type of the content they shall compress. In known approaches, a “blind” signal transformation is performed, by which the multi-channel signal is decomposed into its signal components that are subsequently quantized and encoded. A disadvantage of such approaches is that the computation of the above-mentioned signal decomposition is computationally demanding, and it is difficult and error prone to find the best suitable and most efficient signal decomposition for a given segment of the audio scene.